JPH04253099A - Continuous voice recognition system - Google Patents

Abstract

PURPOSE:To provide the continuous voice recognition system which can recog nize diverse natural sentences at the time of voice recognition and output mean ing expressions. CONSTITUTION:A voice input part 101 analyzes an input voice and outputs the result to a continuous voice recognition part 104. A network storage part 102 is stored with sentences or word strings to be recognized. The continuous voice recognition part 104 selects a word string by matching the pattern of the analytic result of the input voice received from the voice input part 101. An intermediate expression generation part 106 generates an intermediate expression from the ID train of an arc inputted as the recognition result from the continuous voice recognition part 104 and the storage contents of a meaning relation storage part 105 and outputs the generated intermediate expression.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a continuous speech recognition system for recognizing continuously uttered speech in automatic interpretation systems, voice question answering (QA) systems, and the like.

0002

2. Description of the Related Art A man-machine interface using voice is easy to use and natural since it does not require training unlike a keyboard, so its practical use is strongly desired. In addition, it is desirable to be able to handle continuous speech, such as conversational sentences, rather than just single-word speech, as this would be easier to use and more natural.

[0003] When such an interface using audio is used for an automatic interpretation system, audio QA system, etc., the main purpose is not to clarify each word of the input audio, but to extract the meaning of the input audio. In other words, it is necessary to understand the meaning of the input speech.

[0004] This is because a question answering system needs to know the meaning of a user's question in order to respond appropriately, and an automatic interpretation system requires the user to know the meaning of the question in order to respond appropriately. This is because it is necessary to clarify the meaning of the input sentence in order to convert it appropriately.

[0005] As a speech recognition method that extracts the meaning of input speech as a natural language without performing syntactic analysis or semantic analysis on the recognition result of input speech, Japanese Patent Application No.
"Continuous speech recognition method" described in No. 72889 (hereinafter referred to as Document 1)
). In Reference 1, the semantic relationships between words in a word string accepted by an automaton for speech recognition are
When the word string is accepted by the automaton, syntactic analysis and semantic analysis are performed on the speech recognition results of the input speech by storing in advance the semantic relationships between the state transition strings to which each word corresponds. We have proposed a speech recognition method that extracts the meaning of input speech as a natural language.

[0006]

[Problem to be Solved by the Invention] However, in the speech recognition method of Document 1, there is no consideration given to representing the tense of the recognized sentence, the type of sentence such as polite sentence, interrogative sentence, etc. in the intermediate representation. There wasn't.

[0007] Furthermore, as a speech recognition method that can handle context-free grammars with the same amount of processing as a method using finite automata, the 1989 Institute of Electronics, Information and Communication Engineers Spring National Conference ``Continuous Speech Recognition Using Extended Transition Networks'' There is a "one-way method" (hereinafter referred to as Document 2) (Kazunaga Yoshida, Takao Watanabe). Speech recognition using an Augmented Transition Network (hereinafter referred to as ATN) can handle (1) context-free grammars by calling subnetworks; (2) Processing that takes history into account is performed using a register and a mechanism that tests the register, and word order freedom, co-occurrence relationships, and dependency can be handled. He has a high ability to describe natural languages such as

[0008] However, the speech recognition method of Document 1 does not consider the case where the recognition network has a subnet as in Document 2.

[0009] Also, in spoken dialogue in Japanese, for example, the subject is often omitted, as in "Are you going to school?" and "I want tickets to Basia's concert." It is necessary to supplement the omitted elements when creating the speech recognition method, but this consideration is not made in the speech recognition method of Document 1.

An object of the present invention is to overcome these drawbacks and provide a continuous speech recognition method that can recognize more natural and diverse sentences during speech recognition and output their meaning expressions. It is in.

[0011]

[Means for Solving the Problems] The continuous speech recognition method according to the present invention includes a first storage means for storing a network related to words expressing the grammar of a continuous speech recognition target, and a standard pattern of words that is combined according to the network. continuous speech recognition means for recognizing continuous speech; second storage means for storing semantic relationships between one word and another word in the network and feature information for each word; and the continuous speech recognition means. and an output means for outputting a recognition result generated from the recognition result and an intermediate representation thereof from the second storage means.

[0012] The continuous speech recognition means recognizes continuous speech by combining standard patterns of predetermined recognition units according to the network and sub-networks called from each network, and the first storage means recognizes continuous speech. A network representing a grammar to be recognized and a sub-network called from each network are stored, and the output means stores the recognition result generated from the continuous speech recognition means and the meaning between a certain word and another word of the second storage means. The present invention is characterized in that an intermediate representation thereof is output from the relationship and feature information for each word in the network and sub-network.

[0013] Also, the present invention is characterized by comprising the second storage means for storing feature information to be added to another word when a certain word is recognized, and the addition means for adding the information to the intermediate expression. do.

Furthermore, when a certain word is recognized, feature information to be added to the word is retrieved from the word dictionary storage section, and the adding means adds the information to the intermediate expression.

[0015] The word dictionary storage unit stores words that do not exist in the network expressing the grammar of the continuous speech recognition target but are absolutely necessary to create an intermediate representation, and the output means stores words that are necessary for creating an intermediate representation. It is characterized by creating and outputting an intermediate representation.

Further, the word dictionary storage unit stores words to be supplemented when a certain specific word in the network expressing the grammar of the continuous speech recognition target is recognized, and the output means stores intermediate words based on the information. It is characterized by creating and outputting expressions.

Furthermore, the word dictionary storage section stores a specific word in the network expressing the grammar of the continuous speech recognition target that should be deleted when a specific word in the network expressing the grammar of the continuous speech recognition target is recognized. It is characterized in that words are stored, and the output means creates and outputs an intermediate expression based on the information.

[0018]

[Operation] Next, the operation of the present invention will be explained.

FIG. 3 is a diagram showing an example of a semantic relationship storage unit in an embodiment of the present invention, FIG. 4 is a diagram showing a first example of a speech recognition network, and FIG. FIG. 6 is a diagram showing an example of the contents of the relational storage section.
is a diagram showing an example of the contents of the semantic relation storage unit for the sub-network in FIG. 4, FIG. 7 is a diagram showing a second example of the speech recognition network, and FIG. 8 is a diagram showing an example of an intermediate representation of the network in FIG. 7. 9 is a diagram showing an example of the contents of the semantic relation storage unit for the network in FIG. 7, FIG. 10 is a diagram showing a third example of the speech recognition network, and FIG. 11 is a diagram showing the semantic relation for the network in FIG. 10. A diagram showing an example of the contents of the storage unit, FIG. 12 is a diagram showing a fourth example of the speech recognition network, FIG. 13 is a diagram showing an example of an intermediate representation of the network in FIG. 12, and FIG. 15 is a diagram showing the fourth example of the speech recognition network. FIG. 16 is a diagram showing a sixth example of a speech recognition network, and FIG. 17 is a diagram showing an example of an intermediate representation of the network in FIG. 16.

For example, assume that the voice recognition network shown in FIG. 16 recognizes the phrase "I want tickets to Basia's concert tomorrow." Here, in FIG. 16, the numbers attached to each node such as "I am" and "Ticket" are called ID numbers. this"
The intermediate expression of the sentence "I want tickets for Basia's concert tomorrow" is shown in Figure 17.Figure 17
The notation in the intermediate representation corresponding to the words in each speech recognition network, such as "I want" and "Asatte", is called CP. Also, the notation that expresses the relationship between each word, such as "beneficiary" and "object", is called CASE. In Figure 17, #style,
#The information given next to each CP, such as the noun semantic feature, is called feature information. #Style is the sentence type, #Tense is the time, #Verb semantic classification is the semantic classification of the verb, #Noun semantic feature is the semantic feature of the noun, #Quantity is the singular and plural, #Positiveness is the politeness of the sentence. represents that Also, each {
The letters and numbers in } indicate the value of each feature information. In the present invention, for example, in FIG. 3, by storing the semantic relationship between a certain word and another word in the network and the feature information for each word in the network, an intermediate expression as shown in FIG. I'm trying to create one. As a result, the tense of the recognized sentence and the type of sentence, such as polite sentences and interrogative sentences, can be expressed in the intermediate representation.

[0021] Also, consider the case where a voice recognition network such as ATN is used as in Document 2. now,
It is assumed that recognition is performed using a main network and subnetworks as shown in FIG. For example, in the main network of Figure 4 (a), at the "sub date and time", Figure 4 (b)
It jumps to the subnetwork shown in . In this case, in the invention of claim 2, for example, the table shown in FIG. 5 for the network of FIG. 4(a) is
) for the networks shown in Figures 6(a) and (b), respectively.
, (c) are prepared. The subnet table is referenced by specifying the subnet in the CP column in the table of FIG. As a result, it is possible to create an intermediate representation as shown in FIG. 17 even from the speech recognition results using a network such as ATN, which has a main network and sub-networks called from the main network.

Furthermore, in the speech recognition network shown in FIG. 7, for example, when the input voice is "I want one ticket," the intermediate expression is as shown in FIG. 8(a),
Since there is only one ticket, the value of the # quantity in the background information of the CP "ticket" is SIN representing the singular number. but,
If the input voice is "I want two tickets", this intermediate representation will be as shown in Figure 8(b), and since there are two tickets, the value of # quantity in the background information of CP "ticket" is multiple. PL represents.

In this case, in the invention of claim 3, since the information of the CP such as "1 ticket" or "2 tickets" determines the background information of the parent CP "ticket", the "parent grant information" shown in FIG. ” column, information to be added to the background information of the parent CP is stored. As a result, even in the case where feature information is added to another word when a certain word is recognized, it becomes possible to create an intermediate representation of the network as shown in FIG. 7.

By the way, for example, among the descriptions in the feature information column of FIG. 3, the values of #noun semantic feature and #verb semantic classification are "I"
This information is unique to words such as "ticket" and "ticket." Therefore, in the invention of claim 4, each feature information is described in a dictionary and this dictionary is referenced to supplement information such as #noun semantic feature and #verb semantic classification. As a result, the table description can be simplified and the storage capacity of the table can be reduced.

Furthermore, in the case of Japanese, for example, the subject of a sentence is often omitted. Assume that a voice saying "I want tickets to Basia's concert tomorrow", which is one of the recognition results of the network in FIG. 10, is input. This intermediate representation must be the same as in Figure 17, but with CP
The word equivalent to “I” does not exist in the recognition results. In this case, in the invention of claim 5, it is necessary to supplement the CP that is omitted in the voice but is necessary in the intermediate expression, so for example, as shown in FIG. 11, the CP in the ID number column is D by default. shall be taken as a thing. As a result, words that are not present in the speech recognition results but are necessary for creating intermediate expressions are supplemented,
An intermediate representation as shown in FIG. 17 can now be created.

Further, suppose that there is a voice recognition network as shown in FIG. 12, for example. Now, let's consider the case where the input sentence is, for example, "I want tickets for Basia." In this case, the meaning is "I want tickets to Basia's Cosart." The word “concert” is omitted. At this time, the intermediate representation becomes as shown in FIG. 13(a).

Further, assume that the input sentence is, for example, "I want Yume no Yuminsha tickets." In this case, the meaning is ``I want tickets to Yume no Yuminsha's play,'' and the word ``geki'' is omitted. At this time, the intermediate representation becomes as shown in FIG. 13(b). In this case, in the invention of claim 6, when a certain specific word is recognized, a different meaning is added for each word. As a result,
When a specific word is recognized, it becomes possible to create a correct intermediate expression by adding appropriate words.

Also, consider the case where there is a voice recognition network as shown in FIG. For example, this network might say, ``I want tickets to Basia's concert.''
It also accepts input sentences such as ``I want tickets to Basia's concert'' where the word ``I am'' is omitted. All of these intermediate representations are as shown in FIG. 13(a).

In this case, if the invention of claim 5 is used to supplement the CP "I" by default, there will be no problem if "I am" is omitted in the speech recognition result, but if "I am" is omitted in the speech recognition result, there will be no problem. ” exists, an incorrect intermediate representation with two trees indicating that the relationship between “I want” and “I” is a beneficiary is created.

Therefore, in the seventh aspect of the invention, when a certain specific word is recognized, a part of the intermediate expression is deleted to create a meaning expression. As a result, a correct intermediate expression can be created even when a word appears and is omitted in the speech recognition result.

[0031]

[Embodiment] Next, an embodiment of the continuous speech recognition system according to the present invention will be described with reference to the drawings.

First, an embodiment of the invention according to claim 1 will be described. FIG. 1 is a block diagram showing one embodiment of the present invention. In FIG. 1, a voice input unit 101 analyzes input voice as a digital signal, obtains a time series of feature vectors, and outputs the results to a continuous voice recognition unit 104. The network storage unit 102 stores sentences or word strings to be recognized during speech recognition. The numbers assigned to the network are the ID numbers of the arcs representing each word in the network. FIG. 16 shows the network storage unit 10 in FIG.
2 shows an example of a recognition network stored in FIG. When such a network can be traced from the beginning to the end, it is assumed that the input voice is accepted by this network. For example, in the case of FIG. 16, input voices such as "I want tickets to Basia's concert tomorrow" and "I want tickets to Van Heeren's concert tomorrow" can be accepted. The word standard pattern storage section stores in advance the standard patterns of sounds of words to be recognized, and the word standard pattern storage section 103 stores standard patterns of words in the network. This standard pattern may be a concatenation of smaller units, such as syllables.

The continuous speech recognition section 104 connects the standard patterns stored in the word standard pattern storage section 103 according to the word string generated by tracing the above network, and calculates the characteristics of the input speech received from the speech input section 101. By performing pattern matching with the time series of vectors, a word string with the highest acoustic similarity to the input speech is selected. The continuous speech recognition unit 104 outputs a string of IDs of arcs of each word representing this word string to the intermediate expression generation unit 106 as a recognition result. An intermediate representation is created and output from the ID string of the arc input to the intermediate representation generation unit 106 and the contents stored in the semantic relationship storage unit 105.

Next, the processing of the intermediate representation generation section 106 will be explained. Suppose that the input voice is "I want tickets to Basia's concert tomorrow." At this time, the voice recognition result is "1, 3, 5, 6, 10, 11, 13, 15, 1 from the ID number of the arc shown in FIG.
The sequence of arc IDs “6, 17” is the intermediate representation generation unit 10.
Passed to 6. The intermediate representation generation unit 106 creates an intermediate representation from the information shown in FIG. 3 and the ID string of the arc. First, Arc I
When searching the table in Figure 3 for the first 1 in the D number column, the parent ID numbers in the ID number 1 column are 17, 18, 19.
It is. Among these, number 17 is in the ID column of the arc, so the parent of the intermediate expression is "want", and the CASE column shows that the relationship is "benefit". Also, from the feature information column, a feature "#noun semantic feature, 111" is added to "I" in the intermediate expression of this sentence. By repeating the above processing, an intermediate representation as shown in FIG. 17 is obtained.

Next, an embodiment of the invention according to claim 2 will be described. An embodiment of the invention of claim 2 creates an intermediate representation by storing semantic information and feature information of words in a speech recognition network and sub-networks called from each network in the semantic relation storage unit 105 in FIG. . FIG. 4(a) is a main network used for speech recognition, and FIG. 4(b), FIG. 4(c), and FIG. 4(d) are subnetworks called from the main network. For example, Figure 4 (
b) is the subnet “date and time” and arc I in the mainnet.
Called at D number 2.

Now, suppose that the input sentence is "I want tickets for Basia's concert tomorrow." Then, the sequence of arc numbers in the voice recognition result is “1, 2 (2), 3, 4 (1), 5, 6, 8, 1” according to FIGS. 4(a) to (d).
0, 11, 12 (1)". Here, for example, "4 (1
)" means that the arc No. 1 of the subnet is called from the arc No. 4 of the main net. In this case, the subnet with the artist name is called from the sub column of the row with ID number 4 in Figure 5. Figure 6(b)
), it can be seen that the CP is "Basia" and the feature information is #Noun Semantic Feature, 111. Also, the ID in Figure 5
In the row with number 4, the parent ID numbers are 11 and 12, and the ID column of the recognition result is "12 (1)," so from Figure 6(c), point to CP "I want", and from the CASE column, select "12 (1)". It can be seen that ``versia'' and ``want'' have a relationship of ownership. By repeating the above processing, an intermediate representation as shown in FIG. 17 is obtained. The other processing is exactly the same as the embodiment of the invention according to claim 1. As a result, an intermediate representation can be created from the result of speech recognition using the network and the subnetworks called from the network.

Next, an embodiment of the invention according to claim 3 will be described. An embodiment of the invention according to claim 3 also stores feature information to be added to other words in the semantic relationship storage unit 105 in FIG. 1, and creates an intermediate expression using this information as well.

For example, assume that the input voice is "I want one ticket." Then, the ID string of the arc as a result of voice recognition becomes "1, 2, 4, 5, 8" in FIG. At this time, it can be seen that the arc number 5 has a CP of "1" from the row of ID number 5 in FIG. 9, and its parent is a "ticket" because there is "2" in the arc ID column. Further, it can be seen from the parent assignment information column that the identity information "#NUMBER, {SIN}" is assigned to the "ticket" serving as the parent CP. As a result of such processing, an intermediate representation as shown in FIG. 8(a) can be created. The other processing is exactly the same as the embodiment of the invention according to claim 1.

Next, an embodiment of the invention according to claim 4 will be described. FIG. 2 is a block diagram showing another embodiment of the present invention. In this embodiment, an intermediate expression is created by adding feature information to each word not only from the semantic relationship storage section 205 but also using the contents of the word dictionary storage section 207. for example,
Since feature information such as #noun semantic features and #verb semantic classification is always the same for each word, word feature information can be stored by using information in the word dictionary storage unit 207 without having to write it in the semantic relationship storage unit 205. Granted as. The other processing is exactly the same as the embodiment of the invention according to claim 1.

Next, an embodiment of the invention according to claim 5 will be described.

For example, suppose that in the network shown in FIG. 10, the input sentence is "I want tickets for Basia's concert tomorrow." Then the sequence of arc ID numbers is “2, 4, 5, 9, 10, 12, 14, 15, 16”
becomes. It is also assumed that the contents of the semantic relationship storage section are as shown in FIG. Here, it is assumed that the row whose ID number is D is unconditionally used for intermediate expression creation regardless of the result of speech recognition. Since the ID number column of the line containing the CP "I" is "D", the contents of this line are also used to create the intermediate representation. As a result, a correct intermediate representation as shown in FIG. 17 can be created. The other processing is exactly the same as the embodiment of the invention according to claim 1.

Next, an embodiment of the invention according to claim 6 will be described.

For example, suppose that in the network shown in FIG. 12, the input sentence is "I want a Basia ticket." Then the sequence of arc ID numbers is “1, 2, 4, 5”
, 7, 8''. Also, the contents of the semantic relationship storage section are shown in Figure 14.
Assume that it is as shown in . Here, in the on-node column, when the arc with the number described in this column exists in the speech recognition result, the contents of this line are used to create an intermediate expression. At this time, there is "2" in the arc ID number string, so
The contents of the line with the ID number "onl" are also used to create the intermediate representation, and as a result, a correct intermediate representation as shown in FIG. 13(a) can be created. The other processing is exactly the same as the embodiment of the invention according to claim 1.

Next, an embodiment of the invention according to claim 7 will be described. FIG. 18 is a diagram showing an example of the contents of the network semantic relationship storage section in FIG. 15, and FIG. 19 is a diagram showing an example of the intermediate representation of the network in FIG. 15.

For example, suppose that in the network shown in FIG. 15, the input sentence is "I want tickets to Basia's concert." Then the arc ID number column is “1”
, 2, 4, 5, 7, 8, 10, 11''. Also assume that the semantic relationship storage unit is as shown in Figure 18.Here, the off node column is the number written in this column. When the arc of is in the speech recognition result, even if the ID in the same line
Even if the number is "D", the contents of this line are not used to create an intermediate representation. Now, the top row of Figure 18 is I
The D number is “D” but there is “1” in the off node column.
Since there is "1" in the arc ID number column of the voice recognition result, the information in this row is not used. As a result, Figure 19
An intermediate representation such as is obtained.

Also assume that the input sentence is "I want tickets to Basia's concert." Then, the sequence of arc ID numbers becomes “2, 4, 5, 7, 8, 10, 1” in FIG.
1". Now, in the top row of FIG. 18, the ID number is "D".
”, so the contents of this line are also used, and as a result,
An intermediate representation as shown in FIG. 19 is obtained. In this way, in this embodiment, even if there are elements that are omitted or omitted in the network, it is possible to correctly create an intermediate representation. The other processing is exactly the same as the embodiment of the invention according to claim 1.

[0047]

[Effects of the Invention] As described above, according to the continuous speech recognition method of the present invention, it is possible to recognize more natural and diverse sentences during speech recognition, and to output their meaning expressions. .

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram showing an embodiment of the present invention.

FIG. 2 is a configuration diagram showing another embodiment of the present invention.

FIG. 3 is a diagram showing an example of the contents of a semantic relationship storage section in FIG. 1;

FIG. 4 is a diagram showing a first example of a speech recognition network.

FIG. 5 is a diagram illustrating an example of the contents of a semantic relationship storage unit for the main network in FIG. 4;

FIG. 6 is a diagram illustrating an example of the contents of a semantic relationship storage unit for subnetworks in FIG. 4;

FIG. 7 is a diagram showing a second example of a speech recognition network.

FIG. 8 is a diagram showing an example of an intermediate representation of the network in FIG. 7;

FIG. 9 is a diagram illustrating an example of the contents of a network semantic relationship storage unit in FIG. 7;

FIG. 10 is a diagram showing a third example of a voice recognition network.

FIG. 11 is a diagram illustrating an example of the contents of a network semantic relationship storage unit in FIG. 10;

FIG. 12 is a diagram illustrating a fourth example of a network for speech recognition.

FIG. 13 is a diagram illustrating an example of an intermediate representation of the network in FIG. 12;

FIG. 14 is a diagram illustrating an example of the contents of a network semantic relationship storage unit in FIG. 12;

FIG. 15 is a diagram showing a fifth example of a voice recognition network.

FIG. 16 is a diagram showing a sixth example of a speech recognition network.

FIG. 17 is a diagram illustrating an example of an intermediate representation of the network in FIG. 16;

FIG. 18 is a diagram illustrating an example of the contents of a network semantic relationship storage unit in FIG. 15;

FIG. 19 is a diagram illustrating an example of an intermediate representation of the network in FIG. 15;

[Explanation of symbols]

101, 201 Audio input section 102, 202 Network storage section 103, 2
03 Word standard pattern storage unit 104, 204
Continuous speech recognition unit 105, 205 Semantic relationship storage unit 106, 206 Intermediate expression generation unit 207
Word dictionary storage section

Claims (7)

[Claims] 1. A first storage means for storing a network related to words expressing a grammar of a continuous speech recognition target;
continuous speech recognition means for recognizing continuous speech by combining standard patterns of words according to the network; and a second device for storing semantic relationships between one word and another word in the network and feature information for each word. 1. A continuous speech recognition method, comprising: a storage means; and an output means for outputting a recognition result generated from the continuous speech recognition means and an intermediate representation thereof from the second storage means. 2. The continuous speech recognition means combines standard patterns of predetermined recognition units according to the network and sub-networks called from each network to recognize continuous speech, and the first storage means recognizes continuous speech. A network representing a grammar to be recognized and a sub-network called from each network are stored, and the output means stores the recognition result generated from the continuous speech recognition means and the meaning between a certain word and another word of the second storage means. 2. The continuous speech recognition method according to claim 1, wherein an intermediate representation thereof is output based on a relationship between the two words and feature information for each word in the network and sub-networks. 3. The second storage means for storing feature information to be added to another word when a certain word is recognized;
3. The continuous speech recognition method according to claim 1, further comprising: adding means for adding the information to the intermediate representation. 4. When a certain word is recognized, feature information to be added to the word is retrieved from a word dictionary storage section, and the adding means adds the information to the intermediate expression. Continuous speech recognition method described in 2 or 3. 5. The word dictionary storage unit stores words that do not exist in the network expressing the grammar of the continuous speech recognition target but are absolutely necessary for creating an intermediate expression, and the output unit stores words that are not present in the network expressing the grammar of the continuous speech recognition target, and the output unit stores words that are necessary to create an intermediate expression. 5. The continuous speech recognition method according to claim 1, wherein an intermediate representation is created and outputted. 6. The word dictionary storage unit stores words to be supplemented when a certain specific word in the network expressing the grammar of the continuous speech recognition target is recognized, and the output unit stores intermediate words based on the information. 6. The continuous speech recognition method according to claim 1, wherein an expression is created and output. 7. The word dictionary storage unit stores a specific word in the network expressing the grammar of the continuous speech recognition target that is to be deleted when a specific word in the network expressing the grammar of the continuous speech recognition target is recognized. 7. The continuous speech recognition system according to claim 1, wherein words are stored, and the output means creates and outputs an intermediate expression based on the information.